Explained: How ChatGPT Algorithm Is Used to Make Deepfakes

Prime Minister Narendra Modi on Friday addressed the issue of the misuse of artificial intelligence, particularly in the creation of "deepfakes".  Addressing journalists at BJP's Diwali Milan programme at the party's headquarters in Delhi, PM Modi said, "During the times of Artificial Intelligence, it is important that technology should be used responsibly." 

He added, "I recently saw a video in which I was seen singing a Garba song. There are many other such videos online."

PM Modi's comment comes days after deepfake videos of Rashmika Mandanna and Katrina Kaif surfaced online.

What are deepfake videos?

Deepfake videos involve the use of synthetic media to replace a person in an existing image or video. It typically uses generative models and advanced techniques in computer vision, often distinct from language models like ChatGPT. 

How is the ChatGPT algorithm used to make deepfake videos?

• Deepfakes are created through a combination of machine learning techniques, particularly using generative models like Generative Adversarial Networks (GANs) and deep neural networks. The first step entails gathering a large dataset of images or videos featuring the target person whose likeness will be used in the deepfake.
• Deepfake technology uses rules and labelled data to break down sentences. These algorithms can understand simple commands, as seen in chatbots and voice assistants. However, they pose a cybersecurity risk through text analysis and sentence structuring, particularly in actions like enquiring about "what" and "when."
• Generative models, such as GANs, are trained on the preprocessed data. GANs consist of a generator and a discriminator. The generator creates synthetic content, and the discriminator evaluates its realism. 
• Once the model is trained, it can be used to generate deepfake content. For videos, the model might need to consider temporal aspects to create realistic facial expressions and movements. ChatGPT might be used to make deepfake videos with speech that sounds just like a real person, making the fake video more believable. Deepfake videos or images may undergo post-processing to enhance realism, adjust lighting, or refine details. 
• To tackle these concerns, it is crucial to create tools for spotting deepfake videos and boost AI literacy among the public. Establishing legal frameworks may be necessary to control the use of deepfake technology and prevent misuse.